JPH06105466B2 - Image pattern recognition method - Google Patents

Description

Description: FIELD OF THE INVENTION The present invention relates to a new recognition method, as well as to non-human recognition and a thinking method based on it.

More specifically, the present invention relates to a method for recognizing patterns such as figures and characters, and more particularly to a method for recognizing patterns such as figures and characters by a new judgment operation.

BACKGROUND OF THE INVENTION Recognition by conventional mechanical or electrical means, i.e.
The judgment was made only by judging whether or not the physical quantities such as the length, the current value, and the voltage value are numerically the same, or the size thereof. In other words, the computer also relies on the binary logic based on the on / off state of the electric current, and due to the theoretical limit of this judgment, human thought, especially operations similar to human judgment, could not be performed. . Further, due to such a limit, even in pattern recognition, recognition cannot be performed unless the pattern, which is originally an analog amount, is finally converted into a digital amount and processed.

Problems to be Solved by the Invention Here, considering the form of human recognition, there are
= A (or A ') is the basis for determining the identity. The determination of the same means not the comparison of the absolute values of the physical quantities as in the conventional recognition by mechanical or electrical means, but the determination of the sameness as a concept. That is, in human thinking operation, for example,
Dogs, horses, and cows are the same as mammals, and men and women are the same as included in the concept of humanity, and so-called categorization is possible. That is, in this categorization, the determination of the identity of A = A (or A ′) is repeated while maintaining the identity at each stage of the category, so that the recognition target can be identified by the known concept system. Moreover, it is possible to recognize the uniqueness of the recognition target itself.

However, it is impossible to judge such identity by conventional mechanical or electrical means. Therefore, even in the conventional pattern recognition, a plurality of features can be formed from a pattern such as a character or a figure by a complicated procedure. In most of the cases, the distance and the like are extracted, that is, a simple measurement of the absolute amount is performed between the extracted and extracted standard pattern feature groups.

Therefore, in the conventional pattern recognition, when the position of the pattern, the rotation angle, or the scaling factor changes, the recognizable pattern position, the rotation angle, or the scaling factor is limited, such as recognizing the same pattern as a different pattern. is there. Also, in order to give the degree of freedom of position and angle so that it can be correctly recognized even if the position and rotation angle of the pattern change, the conventional pattern recognition method is a method that uses digital feature calculation. There is a problem that the amount of calculation processing for correcting movement and rotation is significantly increased.

Therefore, it is an object of the present invention to provide a novel recognition method capable of determining the conceptual identity of a recognition target at each stage of a category, and also in pattern recognition, the position and rotation of the pattern are recognized. Not affected by the angle,
Moreover, it is an object of the present invention to provide a pattern recognition method that does not require complicated processing.

Means for Solving the Problems The inventor of the present invention has variously studied for the above-mentioned purpose,
As artificial intelligence, we have succeeded in providing a novel recognition method that can judge the identity as the above concept for each category stage.

That is, according to the present invention, there is provided a method of recognizing a recognition target image pattern based on a system of one concept, wherein the system of the concept is composed of a group of concepts that can be replaced with vibration frequencies in a one-to-one manner. The recognized image pattern is converted into a physical system having a non-linear vibration characteristic that belongs to the concept system and has a frequency corresponding to the concept unique to the recognized image pattern, and the physical system is excited to By comparing the vibration frequency converged by the "pull-in" effect with the vibration frequency peculiar to the concept of general knowledge,
An image pattern recognition method is provided which is characterized by determining identity.

Here, "pull-in" means that when two non-linear vibration systems having different frequencies coexist, these two non-linear vibration systems vibrate at the same frequency after a predetermined period due to the similarity of these frequencies. A phenomenon. For the "pull-in" effect of this nonlinear oscillator, see "Nonliner Oscil
ations ", AHNayfeh and DTMook, John Wiley & Sons
See 1979.

The principle of the invention The present inventor has conceived to express a pattern with a single numerical value. It was found that the ratio (2δΓ / δC) of twice the total length δΓ of the contour line of the pattern and the total length δC of the convex envelope of the recognized pattern takes a unique value for each pattern. Here, the convex envelope is an envelope connecting only the corners protruding outside the contour line of the pattern. Therefore, the envelope has no indentation. Some pattern recognition methods according to the present invention have been made based on such findings.

That is, an index that is the ratio of the total length of the contour line of the recognized pattern and the total length of the convex envelope of the recognized pattern is obtained, and the index of the stored pattern data is collated with the index, and the difference is Categorization of the recognized pattern can be performed from the pattern data within the ΔΩ, which means the stored index.

Now, as an example of four patterns of humans, apes, quadrupeds, and birds as shown in FIGS. 1 (a), (b), (c), and (d), each of the above patterns The indicators utilized in the method of the present invention are 3.483, 3.484, 3.
7 and 2.8. If you compare them, the difference ΔΩ is ΔΩ
⁽¹⁾ If it is within (= 0.1), the difference ΔΩ between human 3.483 and ape 3.484 is within the range ΔΩ ⁽¹⁾ (= 0.1), and if they belong to the same category, for example, a diped Can say On the other hand, since the difference ΔΩ between tetrapods and birds is not within 0.1, they do not belong to the same category range. However, if the size of the difference ΔΩ is changed to, for example, ΔΩ ⁽⁰⁾ (= 0.5), the difference between humans, apes and quadrupeds falls within 0.5, and belongs to the same category, for example, animals. . On the contrary, if the difference ΔΩ is reduced to, for example, ΔΩ ⁽²⁾ (= 0.0005),
The difference between humans and apes disappears within that range and both belong to different categories, eg humans and apes. Therefore, if the difference ΔΩ is made small and the recognized pattern is stored and accumulated, the category in a very narrow range, that is, the name of the category itself can be known.

If you look at each of the categories of “animal”, “bipod”, “human”, “ape”, “quadruped”, and “bird” in a tree structure or a hierarchical structure, “animal” is
It is a broad concept that includes all of "human,""ape,""quadruped," and "bird," and can be compared to the "system of concepts.""Biped" is included in the concept of "animal",
It is a narrower concept than "animal," but "human"
Since it is a broader concept than "Apes", "Higher-level concept groups"
Can be compared to And also "quadruped"
Since it is conceptualized at the same level as "biped",
It can be compared to the "top concept group.""Human",
"Ape" is included in the concept of "bipeds" and is a narrower concept than "bipeds" and can be praised as "subordinate concept group". "Birds" can be regarded as conceptualization at the same level as "humans" and "apes," so they can be compared to "subordinate concept groups."

Therefore, as in the method described above, by obtaining the index of the pattern to be recognized and appropriately selecting the range of the difference ΔΩ from the index, the category to which the stored index within the range of the difference ΔΩ belongs is known. By doing so, the pattern to be recognized can be defined as "system of concepts", "superordinate concept group",
The concept can be classified into categories based on various concepts including "subordinate concept groups" and "name".

The index used in the above-described recognition method can be likened to the recognition cell having a nonlinear characteristic that constitutes the visual center of the human cerebrum, in other words, the vibration frequency of the nonlinear oscillator system. And this non-linear oscillator can mention the van der Pol oscillator cell mentioned later as an example.

Therefore, when carrying out the above-described recognition method, instead of actually obtaining twice the total length δΓ of the contour line of the pattern and the total length δC of the convex envelope of the pattern to be recognized and calculating the ratio, that is, the index. , With the following configuration, the value corresponding to the index is directly obtained.

That is, a van der Pole oscillator circuit is formed by arranging van der Pole oscillator cells in a two-dimensional lattice and connecting adjacent oscillator cells, and the two-dimensional bit of the recognized pattern in the van der Pole oscillator circuit is formed. When a plurality of transducer cells corresponding to bits connected to each other in a pattern are selected and excited, a "pull-in" effect of vibrating at one frequency depending only on the shape of the two-dimensional bit pattern of the recognized pattern occurs. . Then, the one frequency corresponds to the index described above.

Here, the "two-dimensional bit pattern" is, for example, a set of 1s and 0s arranged on a square lattice point shown in FIG. 2 (black circles indicate 1 and white circles indicate 0). Further, the term "concatenation" as used herein means that at least one of the eight bits adjacent to each bit having a value of 1 has the same value of 1 as shown in FIG.

Therefore, not only the pattern but also the unknown object to be recognized is converted into a non-linear vibration system such as the van der Pol oscillator circuit described above, and the non-linear vibration corresponding to the vibration of the non-linear vibration system and any concept belonging to the known concept system is converted. It is possible to determine the identity between the concept unique to the object to be recognized and the above arbitrary concept by combining with the vibration of the vibration system and whether or not these vibrations cause "pull-in".

Similarly, when an unknown object to be recognized (for example, a pattern) is converted into a non-linear vibration system (for example, a van der Pol oscillator circuit) and the non-linear vibration system is excited at a vibration frequency peculiar to a known concept, the non-linear vibration If the system causes "pull-in", both can be said to be the same concept.

Furthermore, an unknown object to be recognized (for example, a pattern) is converted into a non-linear vibration system (for example, a van der Pol oscillator circuit), and the non-linear vibration system is excited. As a result, the non-linear vibration system causes "pull-in". If the converged stable vibration frequency is compared with the vibration frequency peculiar to a known concept (for example, a pattern) and they match, both can be said to be the same concept.

That is, among the van der Pol oscillator cells arranged in a two-dimensional lattice as shown in FIG. 4, the oscillator cells corresponding to the value 1 of the bit pattern are selected, and the selected oscillator cells are placed on the bit pattern. When a natural frequency proportional to the number of adjacent bits of value 1 is given, and an interaction is given between adjacent oscillator cells, for example, in the form of a resistance proportional to the speed difference, the operation is "pull-in". Due to the effect, each oscillator cell vibrates at the same frequency after about 10 cycles from the start of the operation. This frequency is measured and output as a frequency corresponding to the bit pattern.

This frequency depends only on the shape of the input bit pattern. When the matrix is composed of vertical and horizontal oscillators as shown in Fig. 4, the pattern is 90 °, 180 °, 2
It does not change when rotating in any of 70 °, when the left and right are swapped, or when moving in parallel.

By processing the central part of the processing calculation by the analog circuit in this way, the calculation operation is not affected by the increase in the degree of freedom such as parallel movement and rotation, and the processing can be performed within 0.1 to 0.2 milliseconds. Also, when storing or transferring patterns, each pattern can be compressed to one frequency, so that the required storage capacity or transfer capacity can be reduced.

Embodiment Hereinafter, an embodiment of an apparatus for carrying out the method according to the present invention will be described with reference to the accompanying drawings.

First, the details of the Van der Pol oscillator circuit described above will be described.

(A) Van der Pol oscillator cell The magnitude of mechanical displacement or displacement of voltage in an electric circuit is represented by x. When x exhibits the operation represented by the following differential equation, it is called a van der Pol oscillator cell.

However, t is a time (second), w _i is a natural frequency, and a and b are scalar parameters.

(B) How to give natural frequencies The natural frequencies of the van der Pol oscillator cells arranged in a two-dimensional lattice are given as follows. When there are n _i bits having the value 1 among the 8 adjacent bits of the number i, the natural frequency w _i of the van der Pol oscillator cell of number i is determined by the following formula.

w _i ² = w ² (1 + cn _i ) ·· (2) However, w is a reference frequency, for example, 2π × 10 ⁵ rad / sec, and c is a constant and has a magnitude of, for example, 0.1.

(C) Regarding how to give the interaction between the oscillator cells: A resistance proportional to the speed difference between the van der Pol oscillator cells i and j corresponding to two adjacent bits having a value of 1. Put in.

For example, the equations of two adjacent oscillator cells A and B among the van der Pol oscillator cells shown in FIG. 5 are as follows.

The same formula holds for the transducer cells C, D, and E.

(D) Entrainment phenomenon If a = 0 in the equation (3), x _A , x _B , x _C, ... Are linear oscillators. In this case, the resonance is w _A = w _B = w _C =
... otherwise it won't happen. On the other hand, in a van der Pol non-linear oscillator in which a is not 0, even if the natural frequencies w are not equal to each other, a phenomenon (retraction) similar to resonance occurs at a frequency between the maximum value and the minimum value of w, and all The oscillator vibrates with the same phase and amplitude. It is known that retraction is particularly likely to occur in the interaction proportional to the difference in speed.

(E) Concatenation of bit patterns In FIG. 6 (a), with respect to the central grid point A, 4 grid points with numbers 1 to 4 are 4 adjacent grid points, and 8 grid points with 1 to 8 are 8 adjacent grid points, 1 to 12 grid points are defined as 12 adjacent grid points. Further, the "connected bit pattern" refers to a pattern in which each bit having the value 1 has at least one adjacent grid point having the value 1.

In FIG. 6 (b), a and b are not connected bits when considering only 4 adjacent lattice points, but are connected when considering 8 adjacent lattice points. Therefore, in the embodiment of the present invention, a bit pattern connected from the standpoint of 8 adjacent lattice points is handled.

(F) Overall operation Therefore, one bit pattern is extracted from one recognized pattern, and the natural frequency w is given to the van der Pol oscillator cell corresponding to the bit pattern according to the rule described in the above (b), The solid interaction described in the item (c) is applied between the adjacent transducer cells to vibrate. The pattern separation ability is improved by using 8 adjacencies as compared with 4 adjacencies, 12 adjacencies as compared with 8 adjacencies (the same applies hereinafter). Then, one frequency f is obtained by drawing in, and compared with the already stored set of frequencies {f ₁ , f ₂ , f ₃ , f ₄ , ... If they are the same (or similar), they are collated.

As described above, the separation ability is improved when the number of adjacent pieces is increased. For example, considering the bit pattern examples shown in FIGS. 7 (a), (b) and (c), if up to 4 neighbors are made to interact as neighbors, (a), (b) and (c) are all Considered the same. However, considering up to 8 neighbors, the interaction becomes as shown in FIGS. 8 (a), (b), and (c), and the bit patterns shown in FIGS. It is distinguished from the bit pattern shown in FIG.

(G) Result of computer simulation FIG. 9 shows the result of calculating the output frequency for some input bit patterns by a computer.
In the figure, the ● mark in the input pattern indicates the bit value 1, and the solid line indicates that there is an interaction between the van der Pol oscillator cells corresponding to the bits. The value of the parameter is a = 0.1, b using the symbols of the above formulas (2) and (3).
= 0.1, D = 0.5, w = 6.0 × 10 ⁵ rad / sec, c = 0.1
Is. It should be noted that these parameters are merely examples.

(H) Embodiment FIG. 10 is an overall configuration diagram of a recognition apparatus for carrying out the pattern recognition method according to the present invention. The recognized pattern is input to the feature extraction unit 10 and converted into a bit pattern. For example, the feature extraction unit 10 includes an imaging device and a binarization circuit that binarizes a video signal from the imaging device. The output of the feature extraction unit 10 is input to the Van der Pol oscillator circuit 12, the natural vibration frequency f _a is input to the verification unit 14. The matching unit 14 includes a data storage unit.
16 comes with, collation unit 14 collates the frequency f _a of the recognition pattern, the frequency f ₁ of different pattern data stored in the data storage unit 16, and f _2, f ₃ · ·, The recognized pattern is specified in an appropriate range from a large category to a small category according to the magnitude of the difference.

FIG. 11 shows the feature extraction unit 10 and the van der Pol oscillator circuit.
12 is a diagram illustrating an interface between the input data storage unit 20 and the bit pattern from the feature extraction unit 10. FIG.
, And each bit of the input data storage section 20 is connected to the van der Pol oscillator circuit 12 via a space switch 22. A frequency measuring unit 24 is attached to the space switch 22. Among these elements, the input data storage unit 20, the space switch 22, and the Van der Pol oscillator circuit 12 each have a cell structure.

The input data storage unit 20 two-dimensionally arranges and stores the bit pattern input from the feature extraction unit 10 via the data input line 26, and outputs it to the space switch 22 as a digital bit signal via the line 28.

In the space switch 22, each van der Pol oscillator cell 30
While receiving the oscillator cell voltage V _out output from the device, the feedback voltage V _in1 and the feedback voltage V _in2 are output to the oscillator of the corresponding van der Pol oscillator circuit according to the bit signal input via the line 28.

Further, the space switch 22 extracts the AC component of the oscillator cell voltage V _out and outputs it as an oscillating voltage to the frequency measuring unit 24 via the line 32. The frequency measuring unit 24 measures the frequency of the AC component and outputs it as a digitized frequency signal 34.

FIG. 12 is a configuration diagram of the space switch 22. The space switch 22 uses the bit signal [Bit
The position (i, j) according to the value 1, 0 of (i, j)]
By opening and closing the five analog switches 36 attached to the transducers (i, j) to output the feedback voltage V _in1 and the feedback voltage V _in2 to the adjacent transducers and the feedback voltage V _in1 from the adjacent transducers. , The feedback voltage V _in2 is input to the vibrator at the position (i, j).

Further, the AC component of the oscillator cell voltage V _out from each oscillator of the Van der Pol oscillator circuit 12 is output as an oscillation voltage to the frequency measuring unit 24 via the line 32.

The feedback voltage V _in1 is the oscillator cell voltage V
It is created by multiplying _out by the weight of the total number of adjacent grid points having a value of 1. In addition, the feedback voltage V _in2
Is produced as the sum of the output voltages of the transducer cells corresponding to adjacent grid points having the value 1.

For example, the transducer cells (i, j), (i, j-1), (i,
When only three (j + 1) have the value 1, the feedback voltage V _in1 and the feedback voltage V _in2 of the transducer cell (i, j) have the following values.

V _in1 = V _out (i, j) × 2 (4) V _in2 = V _out (i, j−1) + V _out (i, j + 1)
(5) FIG. 13 is a block diagram showing a specific configuration of one van der Pol oscillator cell 30 of the Van der Pol oscillator circuit 12. V _in2 is sign-inverted by the inverter 40, is added to V _in1 by the adder 42, is further differentiated by the differentiator 44, and is (3) via the divider 46 and another adder 48.
The voltage corresponding to the interaction term of the equation is input to the Esaki diode oscillator circuit 50. And V _in1 is further
It is inputted to the Esaki diode oscillation circuit 50 as a part of the natural frequency of the equation (2) via another divider 52 and the adder 48.

FIG. 14 is a circuit diagram of the Esaki diode oscillator circuit 50. This circuit is a well-known LC oscillation circuit utilizing the negative resistance characteristic of the Esaki diode 60. Esaki diode
Since 60 has a negative resistance characteristic in which the current is approximately a cubic function of voltage, the output voltage of the oscillator circuit using it follows the equation (1) of the Van der Pol oscillator cell. Variable resistor 62
Has a function of adjusting the negative resistance characteristic of the Esaki diode 60 and increasing or decreasing the amplitude of the oscillator cell voltage V _out . Variable resistor 6
The change in the resistance value of 2 corresponds to changing the values of the parameters a and b in the equation (1). Esaki diode
Assuming that the voltage across 60 is Ved and the voltage of the DC constant voltage power supply 64 is Vc, the amount indicated by x in the equation (1) has the following value in this case.

x = Ved−Vc (6) The operational amplifier 66 inverts Ved to obtain −Ved, and the operational amplifier 68 sums it with Vc to invert the sign. As a result, the oscillator cell voltage V _out becomes the value of the expression (6). In addition,
The resistor 70 is a feedback resistor for inputting the feedback input 54, and the feedback input 54 causes the oscillation circuit to perform the operation shown by the equation (3).

An example of recognizing an image pattern has been described above, but the present invention is not limited to image pattern recognition, and can recognize other recognition targets. If it is explained in the light of the above-mentioned embodiment, the recognition target other than the image pattern, for example,
It can be easily understood that the sound can be recognized by the above-described embodiment once the sound is converted into the image. Therefore, in the present invention, the “recognition target image pattern” should be understood to include an image pattern obtained by converting a recognition target other than the image pattern.

EFFECTS OF THE INVENTION As is clear from the above, in the recognition method according to the present invention,
It is possible to discriminate the features of the object to be recognized by a concept, that is, a group of concepts that can replace the frequency of non-linear vibration such as a pattern, an analog value, and a color, in a one-to-one manner, and a pattern recognition method using this. According to this, the pattern can be recognized without being affected by the position and the rotation angle of the pattern and without requiring complicated processing.

[Brief description of drawings]

1 (a), (b), (c) and (d) are diagrams illustrating the principle of the pattern recognition method according to the present invention, and FIG. 2 is a conceptual diagram showing an input pattern, and FIG. FIG. 4 is an explanatory diagram of adjacent bits, FIG. 4 is a configuration diagram showing an arrangement of transducer cells of a van der Pol oscillator circuit, FIG. 5 is an explanatory diagram of coupling of transducer cells, and FIG. ) And (b) are explanatory views of the concept of adjacent bits, FIGS. 7 (a), (b), (c) and FIG. 8 (a),
(B) and (c) are explanatory views of the connection of transducer cells, FIG. 9 is an explanatory view showing the result of computer simulation, and FIG. 10 is the entire apparatus for implementing the pattern recognition method according to the present invention. Configuration diagram, FIG. 11 is a diagram showing a configuration of an interface between a bit pattern input section and a van der Pol oscillator circuit, FIG. 12 is a configuration diagram of a space switch, FIG. FIG. 14 is a configuration diagram showing an Esaki diode oscillation circuit which is a main part of a transducer cell. [Main reference numbers] 10 …… Feature extraction unit, 12 …… Van der Paul oscillator circuit, 14 …… Verification unit, 16 …… Data storage unit, 20 …… Input data storage unit, 22 …… Space switch, 24 ...... Frequency measurement section

 ─────────────────────────────────────────────────── ─── Continuation of front page (72) Inventor Kenichi Hibino 1-11-15 Nakamachi, Odawara-shi, Kanagawa (72) Inventor Hiroshi Shimizu 8-4 Yonbancho, Chiyoda-ku, Tokyo No. 608 Yonbancho Housing

Claims (1)

[Claims] 1. A method of recognizing an image pattern to be recognized based on a system of one concept, wherein the system of the concept is composed of a group of concepts that can be replaced with a vibration frequency in a one-to-one manner. The recognition target image pattern is converted into a physical system having a non-linear vibration characteristic that belongs to the concept system and has a frequency corresponding to the concept unique to the recognition target image pattern, and the physical system is excited to "pull in". An image pattern recognition method characterized by comparing the vibration frequency converged by the effect with a vibration frequency peculiar to the concept of general knowledge to judge the identity.